Subject: Re: sleep, usleep, and nanosleep vs. pthreads?
To: Florian Stoehr <firstname.lastname@example.org>
From: Steven M. Bellovin <email@example.com>
Date: 09/17/2005 18:58:20
In message <Pine.NEB.firstname.lastname@example.org>, Florian Stoehr write
>On Sat, 17 Sep 2005, Steven M. Bellovin wrote:
>> I'm trying to get gtkpod working on NetBSD. When I ran it, it went
>> into an CPU-chewing loop in the following code:
>> /* keep space_ipod_free/used updated in regular intervals */
>> static gpointer th_space_thread (gpointer gp)
>> for (;;)
>> usleep (SPACE_TIMEOUT*1000);
>> if (!space_uptodate) th_space_update ();
>> Contrary to the usleep call -- SPACE_TIMEOUT is 4000, so it should have
>> been a 4-second sleep -- it seemed to return either immediately or
>> virtually immediately. Changing it to use nanosleep didn't help.
>> However, using sleep() made it work.
>> I'm puzzled. However, I know nothing of pthreads, so I don't know if
>> some special invocation should have been used by gtkpod to make that
>> work. The problem with usleep occurs on 2.0 and -current; the fix
>> works on -current and hasn't been tried yet on 2.0.
>> --Steven M. Bellovin, http://www.cs.columbia.edu/~smb
>>From the manpage of usleep(3):
The microseconds argument must be less than 1,000,000. If the value
>of microseconds is 0, then the call has no effect.
>It should set errno to EINVAL then.
Clearly pilot error on my part in not seeing that (and a worse error on
the part of the original programmer for doing it...) Any thoughts on
why nanosleep() failed? (Hmm -- maybe I should try it again, and make
sure I got it right...)
--Steven M. Bellovin, http://www.cs.columbia.edu/~smb